The present invention relates to a method for estimating the signal-to-noise ratio for packet transmission and reception systems of signals based on M-DPSK modulations, in particular for systems based on receivers of the non-coherent type with differential demodulator, and an apparatus thereof.

In the present communication systems using adaptive modulations for ensuring a certain quality of service QoS thus maximizing the spectral efficiency may be advantageous. The key idea of the adaptive modulation is reacting to the changes of the channel conditions by using strong modulation schemes in the case of bad channel conditions and employing less strong modulation schemes in the case of good channel conditions for increasing the transmission speed. The adaptive modulations may be employed both in single carrier systems and in multiple carrier systems, in both cases a reliable estimation of the channel conditions is needed, for example an estimation of the signal-to-noise ratio (SNR), in order to choose the modulation to be employed.

The apparatuses for estimating the signal-to-noise ratio or SNR estimators may be divided in two categories: "data aided" estimators and "not data aided" estimators i.e. the estimators acting on a known data sequence and those acting on an unknown data sequence. The present data aided SNR estimators applied to non-coherent receivers at the input of a differential demodulator do not allow good estimations of the signal-to-noise ratio to be obtained in the presence of impairments that cause a progressive phase shifting of the received constellation symbols. An example of such impairments is the presence of carrier frequency offsets, between transmitter and receiver, on the carrier frequency. On the other hand, the present data aided SNR estimators applied to non-coherent receivers at the output of a differential demodulator are stronger against those impairments, as the frequency offsets but more sensitive to the noise.

The document of Guerreri L. et al. "LLR-Based Bit-Loading Algorithm for the turbo Coded HomePlug AV", GLOBECOM 2007: IEEE GLOBAL TELECOMMUNICATIONS CONFERENCE, 1 November 2007 (2007-11-01), pages 140-145, XP031195962 Piscataway, NJ, USA, ISBN: 978-1-4244-1042-2 discloses a method for estimating the signal-to-noise ratio for a packet transmission and reception system of signals having a known data sequence by means of a M-PSK modulation with at least one carrier. The system comprises the packet transmission of the signal with a sequence of N known symbols, with N positive integer number and the transmission comprises a M-PSK modulation of the signal to transmit by means of a M-PSK mapper, the transmission of the M-PSK modulated signal through a channel having constant gain over all the N symbols and in presence of noise with null average, and the reception of the signal at the output of the channel.

In view of the state of the art, it is an object of the present invention to provide a method for estimating the signal-to-noise ratio for packet transmission and reception systems of signals based on M-DPSK modulations which allows good estimation values to be obtained even in the presence of impairments that cause a progressive phase shifting of the received constellation symbols. An example of such impairments is the pretence of a carrier frequency offset. In accordance with the present invention, the object is achieved by a method for estimating the signal-to-noise ratio for a packet transmission and reception system of signals having a known data sequence by means of a M-DPSK modulation with at least one carrier, said system comprising means for packet transmission of the signal with a sequence of N known symbols, with N positive integer number, characterized in that said transmission comprising a M-DPSK modulation of the signal to transmit by means of a M-PSK mapper and a differential block, the transmission of the M-DPSK modulated signal through a channel having constant gain over all the N symbols and in presence of noise with null average, and the reception of the signal at the output of the channel, said method comprises the estimation of the signal-to-noise ratio of the received signal with the division of the N known symbols and of the N samples of the signal at the output of the channel into B blocks of L length with B and L positive integer numbers and B greater than one and wherein B is expressed by $B=\frac{N-L}{L-O}+1\ge \frac{N}{L}$ wherein O indicates the overlapping factor of consecutive blocks having length L and the calculation of the estimation of the signal-to-noise ratio is obtained by means of the equation: $\mathit{SNR}=\frac{\frac{\left(L-1\right)}{B\cdot L}{\displaystyle \sum _{b=0}^{B-1}}|{\frac{1}{L}{\displaystyle \sum _{k=l_b}^{l_b+L-1}}{r}_k{a}_k^{*}|}^2}{\frac{1}{B}{\displaystyle \sum _{b=0}^{B-1}}\frac{1}{L}{\displaystyle \sum _{k=l_b}^{l_b+L-1}}{\left|r_k\right|}^2-\frac{1}{B}{\displaystyle \sum _{b=0}^{B-1}}{\left|\frac{1}{L}{\displaystyle \sum _{k=l_b}^{l_b+L-1}}{r}_k{a}_k^{*}\right|}^2}-\frac{1}{L}$ wherein l_b= b · (L - O) + l where l is the index denoting the position of the first known symbol of the sequence of length N in the packet, r_k is the sample of the received signal at the output of the channel correspondent to the known transmitted symbol, a_k is the M-DPSK modulated known transmitted symbol, $a_k^{*}$ is the complex conjugated of the M-DPSK modulated known transmitted symbol and SNR indicates the estimation of the signal-to-noise ratio.

Again in accordance with the present invention, an apparatus for estimating the signal-to-noise ratio may be provided as set forth in the claim 6.

The following description and drawings illustrate embodiments of the invention that comprise the features of the independent claims as well as other embodiments of related methods and apparatuses that do not comprise all the features of the independent claims but are useful for better understanding the claimed invention.

The features and advantages of the present invention will be apparent from the following detailed description of a practical embodiment thereof, shown by way of non-limiting example in the accompanying drawings.

• Figure 1 is the equivalent base-band scheme of a pass-band packet transmission and reception system of the single carrier type with M-DPSK modulator and demodulator and apparatus for estimating the signal-to-noise ratio in accordance with the present invention;
• Figure 2 is a graph of the estimated signal-to-noise ratio SNRs versus the exact value SNRe of an SNR estimator in accordance with the known art;
• Figure 3 is a graph of the mean square error MSE versus the exact value of the signal-to-noise ratio SNRe of the estimator in accordance with the known art;
• Figure 4 is a graph of the estimated signal-to-noise ratio SNRs versus the exact value SNRe of the SNR estimator in accordance with the first embodiment ;
• Figure 5 is a graph of the mean square error MSE versus the value of the signal-to-noise ratio SNRe of the estimator in accordance with the first embodiment ;
• Figure 6 is a graph of the estimated signal-to-noise ratio SNRs versus the exact value SNRe of the SNR estimator in accordance with the first and second embodiments for some combinations of B and L;
• Figure 7 is a graph of the mean square error MSE versus the value of the signal-to-noise ratio SNRe of the SNR estimator in accordance with the first and second embodiments for some combinations of B and L;
• Figure 8 is a graph of the estimated signal-to-noise ratio SNRs versus the exact value SNRe of the SNR estimator in accordance with the fourth embodiment for some combinations of B and L and O;
• Figure 9 is a graph of the mean square error MSE versus the value of the signal-to-noise ratio SNRe of the SNR estimator in accordance with the fourth embodiment for some combinations of B and L and O;
• Figure 10 is the equivalent base-band scheme of a pass-band packet transmission and reception multiple carrier system with M-DPSK modulator and demodulator and apparatus for estimating the signal-to-noise ratio in accordance with the present invention;
• Figure 11 is a graph of the estimated signal-to-noise ratio SNRs versus the exact value SNRe of the SNR estimator in accordance with the second embodiment for a multiple carrier transmission and reception system;
• Figure 12 is a graph of the mean square error MSE versus the value of the signal-to-noise ratio SNRe of the SNR estimator in accordance with the second embodiment for a multiple carrier transmission and reception system.

Figure 1 shows the equivalent base-band scheme of a pass-band packet transmission and reception system of signals with a single carrier. The M-DPSK modulator 1 comprises an M-PSK bit mapper 2 where M is the modulation order, i.e. the number of possible constellation symbols. The mapper comprises a serial-to-parallel converter followed by a map transforming a sequence of log₂M bit c_i into corresponding constellation points with values $a_k^{*}$ where: $$a_k^{ʹ}=e^{{\mathit{j\phi }}_k}=e^{j\frac{2\pi }{M}k},{}_{}k=0,\dots ,M-1$$ and the modulation period T is given by T = T_c log₂ M where T_c is the information bit period. In the bit mapper comprising the M-PSK mapper 2 and the differential block 3, the phase δ_k associated with the symbol transmitted at the kT instant is the same as the phase transmitted at the preceding instant (k-1)T, i.e. δ_k-1, with a phase increase ϕ_k which may assume one of the M values belonging to the range [0, 2π/M,...2π(M-1)lM]. Therefore, in the M-PSK bit mapper, the phase transmitted at the kT instant is given by: δ_k = δ _k-1 + ϕ_k . The modulated symbol at the output of the differential block 3 is a_k = e^jδ_k. The receiver retrieves the transmitted data using the phase difference between consecutive samples without needing to retrieve the phase offset introduced by the channel and the phase offset between transmitter and receiver on the carrier frequency.

The symbol a_k is associated with a limited-band h_T(t) filter 4 so that at the output of filter 4 there is the signal $s\left(t\right)={\displaystyle \sum _{i=-\infty }^{+\infty }}{a}_i{h}_T\left(t-\mathit{iT}\right)$ where the index k has been replaced by the index i; said signal is at the input of the transmission channel 5. Assuming that the transmission channel 5 is a channel which is flat on the signal band and time-invariant with a constant gain g_ch(t) = G, the signal at the output of the channel 5 still has a PSK structure; noise is added to the signal s(t), preferably an additive white Gaussian noise w(t) (AWGN) having a null average with double-sided spectral power density of N₀/2.

Indicating the convolution by the symbol ⊗, and the frequency and phase offset on the carrier frequency between the transmitter and the receiver by Δf and Δϑ, the signal x(t) at the output of the channel 5 and at the input of the receiver 6 is given by: $$x\left(t\right)=\left(s\otimes g_{\mathit{ch}}\left(t\right)+w\left(t\right)\right)\cdot e^{j\left(2\mathrm{\pi \Delta }\mathit{ft}+\mathrm{\Delta }\vartheta \right)}=G\cdot s\left(t\right)\cdot e^{j\left(2\mathrm{\pi \Delta }\mathit{ft}+\mathrm{\Delta }\vartheta \right)}+\tilde{w}\left(t\right),$$ where $$\tilde{w}\left(t\right)=w\left(t\right)\cdot e^{j\left(2\pi \Delta \mathit{ft}+\mathrm{\Delta }\vartheta \right)}\mathrm{.}$$

Assuming that the frequency offset Δf is sufficiently small and considering the filter 6 of the receiver with a function h_R(t) at the output of the filter, the following equation is given $$r\left(t\right)=x\otimes h_R\left(t\right)\cong e^{j\left(2\mathrm{\pi \Delta }\mathit{ft}+\mathrm{\Delta }\mathit{\vartheta }\right)}{\displaystyle \sum _{i=-\infty }^{+\infty }}{a}_{i}h\left(t-\mathit{iT}\right)+n\left(t\right),$$
where $$h\left(t\right)=G\cdot \left(h_T\otimes h_R\left(t\right)\right)\mathrm{and}n\left(t\right)=\tilde{w}\otimes h_R\left(\mathit{t}\right)\mathrm{.}$$

At the sampling instant kT + t₀ where t₀ is the sampling phase the following equation is given $$r_k=G\cdot \left(e^{j\left[2\pi \mathrm{\Delta }f\left(\mathit{kT}+t_0\right)+\mathrm{\Delta }\mathit{\vartheta }\right]}{\displaystyle \sum _{i=-\infty }^{+\infty }}{a}_i{h}_{k-i}\right)+n_k,$$
where $$r_k=r\left(\mathit{kT}+t_0\right),h_k=h\left(\mathit{kT}+t_0\right)\mathrm{and}{n}_k=n\left(\mathit{kT}+t_0\right)\mathrm{.}$$

Assuming that the function h_k = h(kT + t₀) satisfies the Nyquist condition: $$h_{k-i}=\{\begin{array}{ccc}1& \mathrm{if}& k=i\\ 0& \mathrm{if}& k\ne i\end{array}$$ the following equation is given $$r_k=G\cdot a_k{e}^{j\left[2\pi \mathrm{\Delta }f\left(\mathit{kT}+t_0\right)+\mathrm{\Delta }\mathit{\vartheta }\right]}+n_k=G\cdot a_k{e}^{j\left[2\pi \mathrm{\Delta }f\mathit{kT}+\mathrm{\Delta }\varphi \right]}+n_k,$$
where $$\mathrm{\Delta }\varphi =2\pi {\mathrm{\Delta }\mathit{ft}}_0+\mathrm{\Delta }\vartheta \mathrm{.}$$

The sampled signal r_k is processed by a M-DPSK differential demodulator module 7 and sent to an inverted bit mapper 8 providing decisions on the received bits q_i.

The demodulator multiplies the sampled signal r_k by its delayed and conjugated version $r_{k-1}^{*}\mathrm{.}$ where the symbol * indicates the complex conjugated, providing the sample z_k given by: $$\begin{array}{cc}{z}_k\hfill & =r_k\cdot r_{k-1}^{*}=\left(G\cdot a_k{e}^{j\left(2\pi \mathrm{\Delta }f\mathit{kT}+\mathrm{\Delta }\varphi \right)}+n_k\right)\cdot \left(G^{*}\cdot a_{k-1}^{*}{e}^{-j\left(2\pi \mathrm{\Delta }f\left(\mathit{k}-1\right)T+\mathrm{\Delta }\varphi \right)}+n_{k-1}^{*}\right)\hfill \\ & ={\left|G\right|}^2{a}_k{a}_{k-1}^{*}{e}^{j2\pi \mathrm{\Delta }f\mathit{T}}+n_k\cdot n_{k-1}^{*}+G\cdot a_k{e}^{j\left(2\pi \mathrm{\Delta }f\mathit{k}T+\mathrm{\Delta }\varphi \right)}\cdot n_{k-1}^{*}+G^{*}\cdot a_{k-1}^{*}{e}^{-j\left(2\pi \mathrm{\Delta }f\left(\mathit{k}-1\right)T+\mathrm{\Delta }\varphi \right)}\cdot n_k\hfill \end{array}$$
where the normalized frequency offset is defined as Δf_n = Δf · T. As seen from the preceding formula, the useful term i.e. the term ${\left|G\right|}^2{a}_k{a}_{k-1}^{*}{e}^{j2\pi \mathrm{\Delta }\mathit{fT}}$ is insensitive to the phase offset and, if Δf_n is sufficiently small, the frequency offset may be neglicted.

The disadvantage of such a differential demodulator is due to the fact that the term bonded to the noise is amplified. In order to avoid said disadvantage, the SNR estimators considered in this invention act on the signal r(t) at the input of the M-DPSK block 7 of the receiver.

Considering an N sequence of samples r_k consecutively received and l as the index denoting the position of the first known symbol of the sequence in the packet, assuming the channel gain G to be constant over the N samples and the number N to be sufficiently great so as to average out the noise out, assumed as a null average, and knowing that |a_k|² = 1 for unitary power PSK constellations, the following equation is given: $$\frac{1}{N}{\displaystyle \sum _{k=l}^{l+N-1}}{\left|r_k\right|}^2\cong {\left|\hat{G}\right|}^2+{\hat{\sigma }}_n^2$$
where the values Ĝ and ${\hat{\mathrm{\sigma }}}_n^2$ denote the channel estimation and the noise variance estimation, respectively. Assuming to have N available known symbols at the beginning of the useful transmission and an N sufficiently great, greater than 8 and preferably equal to 32, such as to allow the noise, assumed as a null average, to be averaged out, the following equation is given: $${\left|\frac{1}{N}{\displaystyle \sum _{k=l}^{l+N-1}}{r}_k{a}_k^{*}\right|}^2\cong {\left|\frac{1}{N}{\displaystyle \sum _{k=l}^{l+N-1}}G\cdot e^{j\left(2\pi \mathrm{\Delta }f\mathit{kT}+\mathrm{\Delta }\varphi \right)}\right|}^2$$ and assuming the channel gain to be constant over all the N symbols and the frequency offset Δf_n to be negligible, i.e. Δf_nl ≅ Δf_n (l + N -1), the following equation is given: $${\left|\frac{1}{N}{\displaystyle \sum _{k=l}^{l+N-1}}{r}_k{a}_k^{*}\right|}^2\cong {\left|\hat{G}\right|}^2$$ representing the estimation of the squared channel gain.

The estimation of the signal-to-noise ratio SNR is determined by: $$\mathit{SNR}=\frac{|{\frac{1}{N}{\displaystyle \sum _{k=l}^{l+N-1}}{r}_k{a}_k^{*}|}^2}{\frac{1}{N}{\displaystyle \sum _{k=l}^{l+N-1}}{\left|r_k\right|}^2-{\left|\frac{1}{N}{\displaystyle \sum _{k=l}^{l+N-1}}{r}_k{a}_k^{*}\right|}^2}$$

In the presence of frequency offsets Δf_n, when the condition Δf_nl ≅ Δf_n(l + N -1) is no longer true and therefore the approximation ${\left|\frac{1}{N}{\displaystyle \sum _{k=l}^{l+N-1}}{r}_k{a}_k^{*}\right|}^2\cong {\left|\hat{G}\right|}^2$ is no longer satisfied, the SNR estimation, indicated as SNV, decreases as shown in the graphs of Figures 2 and 3 relating to the estimated signal-to-noise ratio SNRs versus the exact value SNRe over various values of Δf_n and to the mean square error MSE against the value SNRe over various values of Δf_n, respectively. As the frequency offset Δf_n increases, the MSE error is noted to dramatically increase and the values of SNRs diverge from SNRe. In accordance with the invention, the N known symbols a_k and the N samples r_k consecutively received and corresponding to the known symbols a_k are divided into B blocks of length L, N = B·L, i.e. the known sequence N of the data packet and the N received samples now consist of B blocks of length L, with B greater than one; specifically, estimator 100 comprises a divider module 101 allowing the N known symbols a_k and the N samples r_k consecutively received and corresponding to the known symbols a_k to be converted into B blocks of length L. The following equation is given: $$\frac{1}{B}{\displaystyle \sum _{b=0}^{B-1}}{\left|\frac{1}{L}{\displaystyle \sum _{k=l_b}^{l_b+L-1}}{r}_k{a}_k^{*}\right|}^2=\frac{1}{B}{\displaystyle \sum _{b=0}^{B-1}}{\left|\frac{1}{L}{\displaystyle \sum _{k=l_b}^{l_b+L-1}}G{e}^{j\left(2\pi \mathrm{\Delta }f\mathit{kT}+\mathrm{\Delta }\varphi \right)}+\frac{1}{L}{\displaystyle \sum _{k=l_b}^{l_b+L-1}}{n}_k{a}_k^{*}\right|}^2$$
where l_b = L·b+l. Assuming the channel gain G to be constant over the L samples and the number L to be sufficiently great, grater than 8, so as to allow the noise, assumed as a null average, to be averaged out, and the condition Δf_nl_b ≅ Δf_n (l_b + L-1) to be satisfied, the following approximation may be carried out: $$\frac{1}{B}{\displaystyle \sum _{b=0}^{B-1}}{\left|\frac{1}{L}{\displaystyle \sum _{k=l_b}^{l_b+L-1}}{r}_k{a}_k^{*}\right|}^2\cong \frac{1}{B}{\displaystyle \sum _{b=0}^{B-1}}{\left|G_{l_b}\right|}^2\cong {\left|\hat{G}\right|}^2$$

The SNR estimator 100 comprises a block 102 allowing the signal-to noise ratio to be calculated by means of the following equation: $$\mathit{SNR}=\frac{\frac{1}{B}{\displaystyle \sum _{b=0}^{B-1}}|{\frac{1}{L}{\displaystyle \sum _{k=l_b}^{l_b+L-1}}{r}_k{a}_k^{*}|}^2}{\frac{1}{N}{\displaystyle \sum _{k=l}^{l+N-1}}{\left|r_k\right|}^2-\frac{1}{B}{\displaystyle \sum _{b=0}^{B-1}}{\left|\frac{1}{L}{\displaystyle \sum _{k=l_b}^{l_b+L-1}}{r}_k{a}_k^{*}\right|}^2}$$

Figure 4 shows a graph of the estimated signal-to noise ratio SNRs versus the exact value SNRe over values of Δf_n being 10^-3 and 2·10^-3 with B = 2 and L = 16 of the SNR estimator 100, indicated as SNVS(B, L), whereas Figure 5 shows a graph of the mean square error MSE against the value SNRe over values of Δf_n being 10^-3 and 2·10^-3 with B = 2 and L = 16 of the SNR estimator; both graphs show the estimator in accordance with the known art for Δf_n = 0, indicated as BOUND. From the values of Figures 2, 3 and Figures 4, 5, the improvement of the performances of the SNR estimator is apparent as compared to the estimator in accordance with the known art.

In accordance with a second embodiment, assuming similarly to the first embodiment the channel gain G to be constant over the L samples and the condition Δf_nl_b ≅ Δf_n(l_b + L -1) to be satisfied, and assuming the number L, differently from the first embodiment, not to be sufficiently great to average the noise out and always with B > 1, the following approximation may be carried out: $$\frac{1}{B}{\displaystyle \sum _{b=0}^{B-1}}{\left|\frac{1}{L}{\displaystyle \sum _{k=l_b}^{l_b+L-1}}{r}_k{a}_k^{*}\right|}^2\cong {\left|\hat{G}\right|}^2+\frac{{\mathrm{\sigma }}_n^2}{L}$$ and therefore the expression of the signal-to-noise ratio SNR calculated from the block 102 of the estimator 100 occurs by means of the following equation: $$\mathit{SNR}=\left(1-\frac{1}{L}\right)\frac{\frac{1}{B}{\displaystyle \sum _{b=0}^{B-1}}|{\frac{1}{L}{\displaystyle \sum _{k=l_b}^{l_b+L-1}}{r}_k{a}_k^{*}|}^2}{\frac{1}{N}{\displaystyle \sum _{k=l}^{l+N-1}}{\left|r_k\right|}^2-\frac{1}{B}{\displaystyle \sum _{b=0}^{B-1}}{\left|\frac{1}{L}{\displaystyle \sum _{k=l_b}^{l_b+L-1}}{r}_k{a}_k^{*}\right|}^2}-\frac{1}{L}=\frac{\left(L-1\right){\displaystyle \sum _{b=0}^{B-1}}|{\frac{1}{L}{\displaystyle \sum _{k=l_b}^{l_b+L-1}}{r}_k{a}_k^{*}|}^2}{{\displaystyle \sum _{k=l}^{l+N-1}}{\left|r_k\right|}^2-L{\displaystyle \sum _{b=0}^{B-1}}{\left|\frac{1}{L}{\displaystyle \sum _{k=l_b}^{l_b+L-1}}{r}_k{a}_k^{*}\right|}^2}-\frac{1}{L}$$

The SNR ratio comprises a weighting factor 1-1/L to weight the signal-to-noise ratio and a corrective term -1/L, it is noted that, if L is sufficiently great, the weighting factor (1-1/L) and the corrective term -1/L may be ignored, thus recovering the equation of the first embodiment of the estimator. In some cases, the corrective term -1/L alone may be ignored.

From the equations relating to the signal-to-noise ratio SNR of the first and second embodiments, it is apparent that for greater values of B the estimator is stronger with respect to the frequency offsets whereas for low values of L the estimator is more sensitive to the noise.

Figure 6 shows a graph of the estimated signal-to noise ratio SNRs versus the exact value SNRe over values of Δf_n being 2·10^-3 with values of B being 2, 4, 8 and respective values of L being 16, 8, 4 of the SNR estimator in accordance with the first and second embodiments, whereas Figure 7 shows a graph of the mean square error MSE versus the value SNRe over values of Δf_n being 2·10^-3 with values of B being 2, 4, 8 and respective values of L being 16, 8, 4 of the SNR estimator 100 in accordance with the first and second embodiments; the estimator in accordance with the first embodiment is indicated as SNVS (B, L), whereas that in accordance with the second embodiment is indicated as RSNVS(B, L). From the values of Figures 6 and 7, the improvement of the RSNVS estimator performances is apparent as compared to the SNVS with low values of L.

In accordance with a third embodiment, the block 101 of the SNR estimator 100, in addition to the division N = B·L, also allows to overlap consecutive blocks having a length L of a factor O always with B > 1. Given a known sequence having a length N, the number of possible divisions is given by $B=\frac{N-L}{L-O}+1\ge \frac{N}{L}\mathrm{.}$ Defining l_b = (L - O)·b + l, the SNR estimator 100 in accordance with the third embodiment allows the signal-to-noise ratio to be calculated with the block 102 by means of the following equation: $$\mathit{SNR}=\frac{\frac{1}{B}{\displaystyle \sum _{b=0}^{B-1}}|{\frac{1}{L}{\displaystyle \sum _{k=l_b}^{l_b+L-1}}{r}_k{a}_k^{*}|}^2}{\frac{1}{B}{\displaystyle \sum _{b=0}^{B-1}}\frac{1}{L}{\displaystyle \sum _{k=l_b}^{l_b+L-1}}{\left|r_k\right|}^2-\frac{1}{B}{\displaystyle \sum _{b=0}^{B-1}}{\left|\frac{1}{L}{\displaystyle \sum _{k=l_b}^{l_b+L-1}}{r}_k{a}_k^{*}\right|}^2}$$

In accordance with a fourth embodiment, considered as a variant of the third embodiment, the block 102 of the estimator 100 comprises, similarly to the second embodiment, a weighting factor 1-1/L to weigh the signal-to-noise ratio and a corrective term -1/L such that the calculation of the signal-to-noise ratio is performed by means of the following equation: $$\mathit{SNR}=\frac{\frac{L-1}{B\cdot L}{\displaystyle \sum _{b=0}^{B-1}}|{\frac{1}{L}{\displaystyle \sum _{k=l_b}^{l_b+L-1}}{r}_k{a}_k^{*}|}^2}{\frac{1}{B}{\displaystyle \sum _{b=0}^{B-1}}\frac{1}{L}{\displaystyle \sum _{k=l_b}^{l_b+L-1}}{\left|r_k\right|}^2-\frac{1}{B}{\displaystyle \sum _{b=0}^{B-1}}{\left|\frac{1}{L}{\displaystyle \sum _{k=l_b}^{l_b+L-1}}{r}_k{a}_k^{*}\right|}^2}-\frac{1}{L}$$

It is noted that, similarly to the second embodiment, if L is sufficiently great, the weighting factor (1-1/L) and the corrective term -1/L may be ignored, thus recovering the equation of the third embodiment of the estimator. In some cases, the corrective term -1/L alone may be ignored.

Once N has been fixed, more combinations of B and L may be obtained with good performances, as shown in Figures 8 and 9 where there are shown a graph of the estimated signal-to-noise ratio SNRs versus the exact value SNRe over values of Δf_n being 2·10^-3 with values of B being 4, 6, 8, 10, 14 and L being 4, 5, 6, 7, 8 and O being 2, 4 of the SNR estimator in accordance with the second and fourth embodiments, and a graph of the mean square error MSE versus the value SNRe over values of Δf_n being 2·10^-3 with values of B being 4, 6, 8, 10, 14 and L being 4, 5, 6, 7, 8 and O being 2, 4 of the SNR estimator in accordance with the second and fourth embodiments, respectively, the estimator in accordance with the second embodiment is indicated as RSNVS(B, L) whereas the estimator in accordance with the fourth embodiment is indicated as RSNVSO(O, B, L).

The means 101 and 102 of the estimator 100 preferably comprise a microprocessor and a memory on which an application software is running thus allowing the division N = B·L and preferably also the overlapping of consecutive blocks having a length L of a factor O as previously specified and the execution of the calculation of the signal-to-noise ratio according to any one of the preceding equations.

The SNR estimator 100 in accordance with the preceding embodiments may also be used for multiple carrier systems, such as the OFDM (Orthogonal Frequency Division Multiplexing) systems. In these transmission systems, there are multiple phenomena that may cause a progressive phase shifting of the received constellation, resulting in effects which are completely similar to the carrier frequency offset in the single carrier systems. It is for this reason that using the estimator in accordance with the invention may also be useful in multiple carrier systems, specifically at the input of the differential demodulator, rather than the estimator in accordance with the known art. The phenomena responsible for the progressive phase shifting of the received constellation include: the carrier frequency offset, the packet synchronization offset and the sampling frequency offset. Each of these phenomena produces a different effect over the various subcarriers of the system and on the OFDM various symbols forming the transmitted packet. For example, the carrier frequency offset, in a pass-band system, produces a phase shifting of the constellation being constant on all the subcarriers of an OFDM symbol but increasing over time for every received OFDM symbol. On the other hand, the packet synchronization offset produces a rotation proportional to the subcarrier index, however remaining constant over all the symbols of the packet. Finally, the sampling frequency offset produces a rotation of the received constellation being different according to both the subcarrier index and the received OFDM symbol.

Given the complexity of the phenomenon, in an OFDM system, a "data aided" SNR estimator may be implemented in different ways, based on the negative effect to be minimized. Specifically, an SNR estimator may act on the frequency in the various carriers or over time between the different transmitted OFDM symbols. However, in any case, the known transmitted sequence of length N and the N received samples may be divided in accordance with the invention in order to reduce the negative effect of the rotation of the constellation in estimating the value of SNR. By way of non-limiting example, we refer to a pass-band OFDM system where every subcarrier is M-DPSK modulated, i.e. the differential is achieved over time, and no recovery of the carrier frequency is achieved during the reception; an OFDM system in accordance with the invention is shown in Figure 10. The M-PSK modulator 200 comprises a serial-to-parallel converter followed by a map which transforms a sequence of log₂M bit c_i into corresponding constellation points, $a_l^{ʹ}$ The M-PSK mapper 200 is followed by a serial-to-parallel converter 201 having N_FFT outputs with values of $a_{n,k}^{ʹ}$ where n and k represent the index of the OFDM symbol and the index of the subcarrier, respectively, and having a duration of T·N_FFT where T = T_c log₂ M, being T_c the information bit period. The M-DPSK modulator comprises N_FFT differential blocks 202 for the output of signals a_n,0... a_n,NFFT-1, being $a_{n,k}=a_{n,k}^{ʹ}\cdot a_{n-1,k},$ at the input of a block 203 to carry out the inverse Fourier transform of said signals and a block 204 for the parallel-to-serial conversion of the signals. The obtained signals are at the input of a channel 205 with a transfer function G(f).

The OFDM systems allow the signal band to be divided into sub-bands in which the transfer function of the channel G(f) may be approximated as almost flat. Under these conditions, the estimator in accordance with the present invention may independently be applied to every OFDM subcarrier. Once the noise w(t) has been added, the passage through the serial-to-parallel converter 206 and the block for applying the Fourier transform 207, in the presence of an offset between the carrier frequency of the transmitter and receiver, the sample relating to the n^th symbol and the h^th subcarrier has the following expression: $$r_{n,h}={\displaystyle \sum _{k=0}^{{\mathit{N}}_{\mathit{FFT}}-1}}{a}_{n,k}{G}_k{\mathrm{sinc}}_{{\mathrm{N}}_{\mathrm{FFT}}}\left(k+\mathrm{\Delta }f\frac{{\mathit{N}}_{\mathit{FFT}}}{F}-h\right)e^{-\mathit{j\pi }\left(k+\mathrm{\Delta }f\frac{{\mathit{N}}_{\mathit{FFT}}}{F}-h\right)\frac{{\mathit{N}}_{\mathit{FFT}}-1}{{\mathit{N}}_{\mathit{FFT}}}}+W_{n,h}$$
where a_n,k and G_k are the transmitted M-DPSK symbol and the complex coefficient of the channel gain on the k^th subcarrier, respectively, W_n,h represents the noise after the discrete Fourier transform (DFT), N_FFT is the number of subcarriers, F=1/T is the sampling frequency and the function $${\mathrm{sinc}}_{\mathrm{N}}\left(x\right)=\frac{1}{N}\cdot \frac{\sin \mathit{\pi x}}{\sin \frac{\pi }{N}x}\mathrm{.}$$

The preceding relation may be written as: $$r_{n,h}=\left(a_{n,h}{G}_n{\mathrm{sinc}}_{{\mathrm{N}}_{\mathrm{FFT}}}\left(\mathrm{\Delta }f\frac{{\mathit{N}}_{\mathit{FFT}}}{F}\right)\right)e^{\mathit{j\pi }\frac{\mathrm{\Delta }f}{F}\frac{{\mathit{N}}_{\mathit{FFT}}-1}{{\mathit{N}}_{\mathit{FFT}}}}+I_{n,h}+W_{n,h}$$
where $$I_{n,h}={\displaystyle \sum _{\begin{array}{c}k=0\\ k\ne h\end{array}}^{{\mathit{N}}_{\mathit{FFT}}-1}}{a}_{n,k}{G}_k{\mathrm{sinc}}_{{\mathrm{N}}_{\mathrm{FFT}}}\left(k+\mathrm{\Delta }f\frac{{\mathit{N}}_{\mathit{FFT}}}{F}-h\right)e^{\mathit{j\pi }\left(k+\mathrm{\Delta }f\frac{{\mathit{N}}_{\mathit{FFT}}}{F}-h\right)\frac{{\mathit{N}}_{\mathit{FFT}}-1}{{\mathit{N}}_{\mathit{FFT}}}}$$ represents the channel interference due to the frequency offset Δf, said term degrades the performances of the OFDM systems and makes the SNR estimators 100 more sensitive to the frequency offset as compared to single carrier transmission and reception systems. The frequency offset determines a phase rotation for the received symbols being constant over all the subcarriers, said phase rotation is proportional to the normalized frequency offset Δf_n with respect to the sampling frequency F, Δ f_n = Δf / F = Δf · T. The samples r_n,0... r_n,NFFT-1 are at the input of the SNR estimators 100 and then at the input of the M-DPSK receivers 208, parallel-to-serial converter 209, inverted mapper 210 in order to have the estimated bits m_i corresponding to the bits c_i at the output.

It is noted that the SNR estimator is applied at the input of the differential demodulator: if the SNR estimator is applied at the output of the differential demodulator there is a greater insensitivity to the frequency offset, the disadvantage being the noise amplification.

In the graphs of Figures 11 and 12, there are shown the estimated signal-to-noise ratio SNRs versus the exact value SNRe over values of Δf_n being 10^-4 and 5·10^-5 of the SNR estimator for a single carrier in accordance with the known art and the second embodiment being B = 4 and L = 8 and the mean square error MSE versus the value SNRe over values of Δf_n being 10^-4 and 5·10^-5 of the SNR estimator in accordance with the known art and the second embodiment being B = 4 and L = 8, respectively; the estimator in accordance with the known art is indicated as SNV whereas the estimator in accordance with the second embodiment is indicated as RSNVS(B, L). Both graphs show the estimator in accordance with the known art for Δf_n = 0, indicated as BOUND.

From the values of Figures 11 and 12, the improvement of the RSNVS estimator performances is apparent as compared to the SNV estimator.

The used pass-band single carrier communication system preferably works at a carrier frequency of 100 kHz and has a system clock having a tolerance of ± 200 ppm or ± 100 ppm and a sampling frequency being F = 10 kHz with the SNR estimator applied at the input of the differential demodulator.

The used pass-band multiple carrier communication system preferably works at a carrier frequency of 250 kHz and has a system clock having a tolerance of ± 40 ppm or ± 20 ppm and a sampling frequency being F = 100 kHz with the SNR estimator applied at the input of the differential demodulator.